Application of Transfer Learning for Breast Cancer Diagnosis based on Three-dimensional Ultrasound Images

Breast cancer represents a significant global health challenge and has emerged as one of the leading threats to human life, particularly among women. However, The process of diagnosing breast cancer is time-consuming, complicated, and susceptible to inaccuracies. The growing incidence of breast cancer highlights the need for enhanced medical imaging, early detection, and innovative treatment strategies to mitigate its impact on public health. In this study, we developed a three-dimensional (3D) deep learning model for the Multiview ultrasound images from 1000 lesions, which were randomly divided into training, validation, and test sets in a 3:1:1 ratio. The diagnostic performance of the proposed 3D model was compared with the 2D model, and the experiment results showed that the 3D model outperformed the two-dimensional (2D) model with AUC 0.946, to 0.875. When combined with ultrasound images, low-aged radiologist whose AUC was increased from 0.755 to 0.860, nearly reached the level of senior radiologist.